| name | lope-implement |
| description | Run a sprint with zero-human swarm orchestration. First select implementation agents and escalation agents, then Lope executes phases without further human input. Use when the user says implement the whole sprint, stay out of the loop, use Claude/OpenCode or specific CLIs for escalations, or wants autonomous sprint backlog completion. |
Lope Implement
lope implement <sprint_doc> is the zero-human wrapper around lope execute.
It exists for one job: select the agent roster once, then run the sprint without asking the human again.
Invocation
Two paths:
- Interactive TTY
lope implement SPRINT.md
Lope asks:
- Which implementation agents to use
- Which escalation agents to use
After that, no human input.
- Agent / CI / non-interactive
lope implement SPRINT.md \
--agents pi,antigravity \
--escalate-to claude,opencode
In non-interactive mode, both --agents and --escalate-to are required. No silent defaults.
Common flags
lope implement SPRINT.md --phase 2 --agents pi --escalate-to claude,opencode
lope implement SPRINT.md --agents pi --escalate-to claude,opencode --gates
lope implement SPRINT.md --agents pi --escalate-to claude,opencode --dry-run
Minimality discipline
For engineering sprints, lope implement injects minimality audit guidance by default: prefer existing code, stdlib/native features, and the smallest safe custom code; avoid duplicate helpers, one-implementation abstractions, broad rewrites, and symptom patches. This is audit guidance unless LOPE_MINIMALITY=enforce is set. Disable with LOPE_MINIMALITY=off. Business/research domains stay off unless explicitly enabled.
Safety model
v1 is a single-writer swarm.
The first --agents entry is the writing lead. Lope does not ask multiple CLIs to edit the same checkout concurrently because that creates patch races. The selected implementation and escalation agents are injected into the prompt and validator pool. Real parallel implementation belongs in a future worktree-backed mode.
Rules
- Do not run without a sprint doc.
- Do not invent agent names. Run
lope team list if unsure.
- In non-interactive mode, always pass
--agents and --escalate-to.
- Prefer
--dry-run before long autonomous runs.
- Use
--interactive to force the roster picker when TTY detection would otherwise treat the run as non-interactive.
- If validators return NEEDS_FIX, let Lope retry. Do not ask the human.
Runtime safety (v0.14.0)
Use --run-timeout for the whole command; --timeout remains a per-provider-call ceiling. Inspect the emitted request plan before large work and prefer compact evidence. Automatic shaping is bounded; use chunking only when the forecast fits --max-calls and --max-chunks. For abandoned work, run lope jobs list and preview lope jobs reap --dry-run. Never use pkill -f, killall, or process-name matching.